mdmPevHC

the need to provide Big Data & Analitica functionality, with learning AI algorithms, ML/DL, to the clinical data needed for research in compliance with current regulations regarding privacy and security

Inspiration Healthcare personnel are the containment barrier and researchers will be the ones to find the treatment to cure and eradicate Covid19

Now more than ever, it is essential to give a powerful Big Data and Artificial Intelligence tools and the fuel that is quality clinical data to the research community

However, the main barrier that has always been faced researchers in Europe to access health data is regulatory compliance in data protection. Without solving these limitations, it becomes difficult to compete with other competitors.

What it does We have developed an efficient, scalable and secure solution that allows healthcare systems to extract personal information from medical records, anonymize them, tokenize them and have valuable medical data in a European private cloud, strictly complying with the European and national data protection standards.

Once the data is stored in the cloud, using governance and data processing processes, researchers will be able to access datasets to carry out all types of scientific studies using Big Data, Analytical and Artificial Intelligence tools.

How we built it Vicomtech's has develop technology for automatic anonymisation of textual data. Currently, you will find content related to our participation in the 2019 edition MEDDOCAN: Medical Document Anonymization shared task and our medical anonimysation demo, HitzalMed. Upon registration, you can use the demo or download the scripts and models that our team used in the challenge, as well as check the documentation on how to use them yourself.

To acces to evauate Hitzalbel Tool: https://snlt.vicomtech.org/hitzalmed

We have embedded inside Raspberry Pi 4 BERT models. They handled occupy about 600-700 MB on disk and 2-3 GB in memory. They can be loaded on CPU. Each CPU can process about 500 words processed in ~ 1s (counting with the overhead of the web service). The model size can be reduced using DistilBERT, at the cost of a small loss in detection. To put it in perspective: "DistilmBERT reaches 92% of Multilingual BERT's performance on XNLI while being twice faster and 25% smaller" (https://github.com/huggingface/transformers/tree/master/examples/distillation)

eTICa group has designed the processes for regulatory compliance, the design of the BERT models embedded within rasberry Pi, an efficient and scalable architecture in the cloud as well as the selection of Artificial Intelligence tools and the algorithms.

Challenges we ran into Design the different technologies used in the project and their integration.

The way to access and load the data from the medical records from any healthcare center in the world, complying with European regulations.

Analyze existing legal problems to ensure that the information uploaded to the private cloud is correct

Choose the architecture in the cloud and the necessary Artificial Intelligence tools to carry out the investigations.

Planning of the implementation of a MPV for October 2020.

Accomplishments that we're proud of We have agreed with different initiatives to collaborate together on a product that does not exist on the market and that will have a very significant impact worldwide.

• Hitzalmed tool. • Raspberry Pi 4 design of the embedded model • MPV design • Very high performance scalable cloud architecture design • Design of the analytical and artificial intelligence layer

What we learned It is essential to create a new useful solution not available today in the market that solves a problem and meets customer expectations

Providing quality, complete and up-to-date medical information is necessary in order to give researchers the most suitable tools.

Knowing how to transmit the solution's advantages is key to involving the key actors in a moment as complicated as the current one.

What's next for X-COV Find support from the Government, EU Institutions and healthcare systems Finance support to develop the MPV Develop a communication plan for the solution so that it can be understood and used by all European health centers Devices Manufacturing Plan Expand the functionality of the solution to meet all the future needs Improve the Platform analytical capabilities to face of future waves of Covid19 and future crises. Put in place the infrastructure

Built With AWS Stack, Raspberry Pi 4, Python, Apache Spark, TensorFlow, Keras